Bayesian Semiparametric Regression for Autoregressive Models with Possible Unit Roots

In this paper we consider bayesian semiparametric regression within the generalized linear model framework. Specifically, we study a class of autoregressive time series where the time trend is incorporated in a nonparametrically way. Estimation and inference where performed through Markov Chain Monte Carlo simulation techniques. Main results show that treating the time trend nonparametrically possible model misspecification and biased results from structural break issues are solved. Empirical applications are conducted using the extended Nelson and Plosser benchmark time series